On Thu, Jul 21, 2011 at 05:28:42PM +0100, Mel Gorman wrote: > Warning: Long post with lots of figures. If you normally drink coffee > and you don't have a cup, get one or you may end up with a case of > keyboard face. [snip] > Overall, having kswapd avoiding writes does improve performance > which is not a surprise. Dave asked "do we even need IO at all from > reclaim?". On NUMA machines, the answer is "yes" unless the VM can > wake the flusher thread to clean a specific node. Great answer, Mel. ;) > When kswapd never > writes, processes can stall for significant periods of time waiting on > flushers to clean the correct pages. If all writing is to be deferred > to flushers, it must ensure that many writes on one node would not > starve requests for cleaning pages on another node. Ok, so that's a direction we need to work towards, then. > I'm currently of the opinion that we should consider merging patches > 1-7 and discuss what is required before merging. It can be tackled > later how the flushers can prioritise writing of pages belonging to > a particular zone before disabling all writes from reclaim. Sounds reasonable to me. Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx _______________________________________________ xfs mailing list xfs@xxxxxxxxxxx http://oss.sgi.com/mailman/listinfo/xfs
